Fix hwloc topology traversal code unable to handle situation where L2 cache is common for the packages

[pawosm01]

Summary

Currently cores within package that share the same L2 cache are grouped together. The current logic behind this assumes that the L2 cache is always at deeper (or the same) level than the package itself. In case when L2 cache is common for all packages (and the packages are at deeper level than L2 cache) the whole of the further topology discovery fails to find any computational units resulting in following assertion:

Assertion failure at kmp_affinity.cpp(715): nActiveThreads == __kmp_avail_proc.
OMP: Error #13: Assertion failure at kmp_affinity.cpp(715).

This patch adds a bit of a logic that prevents such situation from occurring.

[AndreyChurbanov]

LGTM

openmp/trunk/runtime/src/kmp_affinity.cpp

static int __kmp_hwloc_process_obj_core_pu(AddrUnsPair *addrPair,
   return 0;
 }
 
 // Check if NUMA node detected below the package,
 // and if tile object is detected and return its depth
 static int __kmp_hwloc_check_numa() {
   hwloc_topology_t &tp = __kmp_hwloc_topology;
   hwloc_obj_t hT, hC, hL, hN, hS; // hwloc objects (pointers to)
-  int depth;
+  int depth, l2cache_depth, package_depth;
 
   // Get some PU
   hT = hwloc_get_obj_by_type(tp, HWLOC_OBJ_PU, 0);
   if (hT == NULL) // something has gone wrong
     return 1;
 
   // check NUMA node below PACKAGE
   hN = hwloc_get_ancestor_obj_by_type(tp, HWLOC_OBJ_NUMANODE, hT);
   hS = hwloc_get_ancestor_obj_by_type(tp, HWLOC_OBJ_PACKAGE, hT);
   KMP_DEBUG_ASSERT(hS != NULL);
   if (hN != NULL && hN->depth > hS->depth) {
     __kmp_numa_detected = TRUE; // socket includes node(s)
     if (__kmp_affinity_gran == affinity_gran_node) {
       __kmp_affinity_gran = affinity_gran_numa;
     }
   }
 
+  package_depth = hwloc_get_type_depth(tp, HWLOC_OBJ_PACKAGE);
+  l2cache_depth = hwloc_get_cache_type_depth(tp, 2, HWLOC_OBJ_CACHE_UNIFIED);
   // check tile, get object by depth because of multiple caches possible
-  depth = hwloc_get_cache_type_depth(tp, 2, HWLOC_OBJ_CACHE_UNIFIED);
+  depth = (l2cache_depth < package_depth) ? package_depth : l2cache_depth;
   hL = hwloc_get_ancestor_obj_by_depth(tp, depth, hT);
   hC = NULL; // not used, but reset it here just in case
   if (hL != NULL &&
       __kmp_hwloc_count_children_by_type(tp, hL, HWLOC_OBJ_CORE, &hC) > 1)
     __kmp_tile_depth = depth; // tile consists of multiple cores
   return 0;
 }